BrontoPoint Contact
Mixed criticality · Platform security · Physical AI

Predictable, secure and accountable platforms for critical functions and AI.

BrontoPoint helps platform, automotive and robotics teams run real-time, security-critical and AI workloads on the same silicon, and prove they behave.

BrontoPoint
Non-deterministic in, predictable out
Vision

When software-defined systems move AI and other best-effort workloads onto the same silicon as functions that must not fail, in vehicles, robots, factories and edge systems, they should be trusted only when their behaviour is measured and verified, not assumed.

Mission

Help teams run critical functions alongside AI and other best-effort workloads on shared platforms, with behaviour they can measure, attribute, control and verify.

The track record behind this, with case studies and published benchmarks, is on Lei Zhou's site.

Measure, attribute, control, verify.

One loop, applied to timing and to trust

We use the same four steps for timing problems and for security problems.

  1. MeasureCollect the evidence

    Timing: Profile latency, jitter and interference between workloads on shared hardware.

    Trust: Integrity measurement: measured boot and recorded hashes of what actually loaded and ran.

  2. AttributeExplain the cause

    Timing: Trace each outlier to its source: scheduler, cache, memory bandwidth, interrupt or driver.

    Trust: Tie every measurement to an identity and a chain of trust, so you know who and what it came from.

  3. ControlAct on the cause

    Timing: Configure, isolate or limit: partitioning, scheduling and quality-of-service on the resources that matter.

    Trust: Enforce policy at the boundary: TEEs, hypervisor isolation, key protection and access control.

  4. VerifyProve the result

    Timing: Show the bound holds under load, with repeatable benchmarks and regression evidence.

    Trust: Remote attestation and tamper-evident records that a third party can check.

Three questions we help answer

Each engagement starts from a question your platform has to answer before it ships, and ends with measurements and a plan you can act on.

Predictable

Mixed criticality and virtualization

Will the control loop still meet its deadline when the AI workload spikes?

  • Hypervisor and GPU virtualization architecture for consolidated platforms
  • Latency measurement and attribution on real-time Linux and RTOS
  • Interference characterization and isolation design

Secure

Full-stack platform security and confidential computing

Can you show the device boots, runs and reports only what it should?

  • Hardware root of trust, secure boot and TEE design reviews
  • BSP and firmware security assessments
  • Threat modelling and security requirements, from device to cloud

Accountable

Physical AI and robotics

When AI shares a platform with control, can you trace what it did and when?

  • Platform readiness reviews for AI on embedded SoCs
  • Benchmarking AI inference against real-time control on shared hardware
  • Tamper-evident execution records for AI agents

How an engagement runs

Remote-first, across time zones.

  1. Conversation

    A short call to understand your platform, your constraints and the question you need answered.

  2. Scoped assessment

    A fixed-scope review with a written report: findings, measurements and a prioritised plan.

  3. Advisory or hands-on

    Ongoing architecture advice, or working alongside your engineers to build the evidence.

Work is under NDA by default, and findings stay with you.

Who you work with

BrontoPoint is the engineering advisory practice of Lei Zhou, a platform architect whose work spans semiconductor, automotive, industrial and AI infrastructure.

Articles, talks and open-source work are at zlhk100.github.io.